Análise de desempenho de aplicações científicas em ambiente de nuvem privada

Abstract

Cloud computing infrastructures are becoming a good alternative to execute scientific applications. The reason of this is that they have a high resources allocation flexibility for a cheap price. For its implementation, there are available many open source tools for the management of these environments. However, virtual environments can suffer of a loss of performance caused by the overhead of the virtualization software. This problem becomes worse when some classes of application have a high communication demand among the processes. In this context, this project has as its objective the evaluation of the performance of private clouds infrastructures. In order to do this analysis, two scientific applications were utilized. These applications have a high computational demand, and low communication among the processes. The evaluation was made considering two private clouds, which were managed by OpenNebula and OpenStack. The results showed a bigger loss on the executions that demanded more communication among the processes. Therefore, it lead to the conclusion that the overload of network virtualization has a bigger impact. The results also showed that the tools have an impact on the performance, where OpenNebula had slight better results.